Give your AI agent hands and eyes on every PDF
Every open, dwell, annotation and form fill on your documents — typed, attributed to a named reader, and ready for your agent to query and act on. Over MCP, REST and webhooks. Apdf is the PDF engagement layer for apps and agents.
Most PDF tools for agents
stop at the file.
Reader MCPs tell your agent what's inside a document. Generator APIs render one and hang up. Neither knows what happens after you hit send — and that's where the deal lives.
Read the file
Extract text, parse tables, OCR a scan. Useful — but the story ends at the last page. What a human did with the document never becomes data.
Render and forget
HTML in, PDF out. The moment the file is delivered, the API's job is done. Whether anyone opened it, read it or walked away — silence.
Close the loop
Your agent builds the PDF, shares it to a named reader, watches every move as typed events — and acts on what it sees. Make → share → watch → act, one endpoint.
You own the signal. We host the surface. MCP is one of three rails — the same data flows over REST and webhooks.
PDF MCP tools for the whole engagement loop.
Every tool below is live on the MCP endpoint and mirrored in the REST API. Names map one-to-one — what your agent calls in chat, your backend can call in code.
PDF tracking links for AI agents
The agent publishes a document and mints a tokenized link per recipient — no reader logins, identity embedded in every event that follows. Links activate, deactivate and revoke on command, even mid-read.
PDF engagement analytics for AI agents
Who opened, which page held their attention and for how long, who returned, where they dropped off — plus every annotation and form submission as structured, attributed data. The agent asks; the answer is already typed.
Automation & job monitoring for agents
The agent inspects your when/filter/then rules and their full execution log — payload, response, duration per firing — and checks on long-running PDF jobs without polling blind.
PDF generation & processing for AI agents
The engagement layer sits on a full PDF engine, so the agent can also build the document it's about to track — generate from HTML, merge, split, compress, OCR, watermark, secure — without a second vendor.
Wakes on the signal.
Acts while it's hot.
Chat is the demo. The production shape is event-driven: reader signals wake your agent, MCP gives it context and hands, and the follow-up happens while the document is still open.
A reader moves
Alex leaves the pricing page after a long dwell. The motion lands as a typed event, attributed to the recipient link.
The signal wakes your agent
A when/filter/then automation matches the event and POSTs the payload to your agent's endpoint. No polling, no cron.
It pulls the context
Over MCP, the agent reads the full session: pages, dwell, return visits, the annotation on the rate line.
It acts while interest is hot
A follow-up drafted around the exact objection, the CRM stage moved, a fresh link minted for the CFO.
Playbooks your agent can run on day one.
Each card is a workflow an agent runs end to end — from Claude, Cursor, a framework of your own, or headless against the REST API.
The proposal chaser
Every morning the agent sweeps active proposals, ranks readers by real attention instead of opens, and flags exactly where each stalled reader stopped.
The objection triager
A reader highlights a line and leaves a comment. The agent wakes, reads the exact quote in context, files the objection in the CRM and pings the deal owner.
The intake processor
Vendor intake runs through PDF forms. On every submission the agent checks the typed field data for completeness, files it, and issues a confirmation document.
The deal-desk factory
One instruction builds the proposal from your HTML template, publishes it as a tracked document, and mints a personalized link for every stakeholder on the deal.
Connect once.
Bring any client.
One OAuth-signed MCP endpoint for chat clients and agent frameworks — and the same tools over REST for everything that doesn't speak MCP yet.
Sign in from your client
Or call it like an API
Every MCP tool has a REST twin, and webhooks push events the moment they happen — so headless agents, cron jobs and your own product all drink from the same stream.
API referenceAutonomy, with guardrails.
An agent acting on live reader data needs boundaries you can see and pull. They're built in, not bolted on.
OAuth-scoped access
Agents sign in over OAuth 2.0 and get exactly the workspace access you grant — revoke a connection and the tools go dark.
Revocable mid-read
Every recipient link can be deactivated at any moment — by you, or by an agent that spots something off.
A full audit trail
Every automation firing is logged with payload, response and duration. What acted, when, and on what signal — always answerable.
EU-hosted, GDPR-native
Documents and engagement data live on EU infrastructure by default.
Never training data
Your documents and your readers' behavior are never used to train models. Period.
No lock-in
Full data export on demand, an open API underneath everything. Leave whenever — with everything.